Assessment of excess risks in case-base studies.

Miettinen and Caro (J Clin Epidemiol 1989; 42: 325-331) [6] put forth principles of non-experimental assessment of excess risks in case-base studies; in such a design, risk difference is assessed by estimating the denominators of the proportions of cases among exposed and comparable unexposed subjects by means of a representative sample from the base population of the study. They provided appropriate formulations for the point estimation of risk differences for various exposure patterns with allowance for covariables by means of stratified analysis. However, in small samples point estimates can be uncertain. In this paper, first, likelihood-based statistics are derived which can be used for interval estimation (and also point estimation and significance testing) of risk differences. The unified approach generalizes to stratified analysis. Second, the procedure is parameterized for inferences about risk ratios using a chi-square function in analogy with the profile likelihood method for full cohort analysis. Third, the approach is extended from the study of a binary exposure variable to a risk function analysis under the Poisson regression model. The suggested estimation procedure is conceptually clear and computationally simple in that the modelling, unlike the models considered by Prentice (Biometrika 1986; 73: 1-11) [3] for the analysis of case-base data, focuses directly on the comparison of risks between exposure categories in the study base and thus involves no covariance between the cases and the base sample.